I tried using Stability AI official web application `` Stable diffusion reimagine '' that can infinitely generate `` similar images '' from images Review

Stability AI, the developer of the image generation AI `` Stable diffusion '', has released a service `` Stable diffusion reimagine '' that generates an `` image similar to the reference image '' simply by specifying a reference image. At the time of writing the article, some functions were available for free, so I tried to see what kind of image I could actually generate.

Stable diffusion reimagine


Stable Diffusion Reimagine -Stability AI

When you access Stable diffusion reimagine, the following screen will be displayed. When you drag and drop an image to the part surrounded by the dotted line on the screen, three 'images similar to the dragged and dropped image' are generated.

This time, I will try using the following image.

Drag and drop the image to the part surrounded by the dotted line.

Then the image generation will start, so wait for a while.

Three images were generated in about 10 seconds after dragging and dropping.

Click on the thumbnails at the bottom of the screen to see the full image of each image.

If you don't like the generated result, click the refresh mark on the thumbnail.

A new image is then generated.

To download the image without an account, click the share button located to the left of 'HD'. You can download high resolution images by clicking 'HD', but you need to create a 'ClipDrop' account.

Then, you can download 'an image that combines one original image (upper left) and three generated images' as follows. A human finger is reflected in the lower left image, but all images reproduce the feature of 'a picture of a doll sitting on a simple background'.

I also tried it with other photos. As a result of entering a human portrait, for some reason it is possible to reproduce a 'smiling woman wearing a straw hat' although she is wearing sunglasses.

Below is the result of specifying a photo of two women as a reference image. Compared to the result of inputting a photo of a single person, the depiction of the arm and face looks rough.

When I entered a picture of '

chilled tanuki ', a mysterious food was generated that left the atmosphere of the original picture.

Enter 'picture of curry in a pot'. It seems that green okra was recognized as an herb.

Landscape photos can also be generated in large quantities. The quality is noticeable when used for posters, etc., but the quality is sufficient for low-resolution images such as blogs.

Stability AI acknowledges that Stable diffusion reimagine may generate anomalous images and shows a willingness to improve the system based on feedback.

in Review,   Web Application, Posted by log1o_hf